AI Video Dubber: put a video into another language with a new voice track

Give it a direct link to a video you host and pick a language. It writes down what is said, translates it, speaks the translation with an AI voice timed to the original lines, and puts that back over the picture. You get a dubbed MP4 and a translated SRT.

Two things to be clear about. It runs on your own OpenAI key, so the transcription, the translation and the voice all land on your OpenAI bill on top of the charge here. And the original audio is gone: this is a replacement voice track, not a mix, and there is no voice cloning and no lip sync.

Input	A direct link to a video file you host
Output	One dubbed MP4 and a translated SRT
Ceiling	One video and one language per run
Account needed	Your own OpenAI API key
Price	$0.12 per minute of source video, rounded up

🌍 What AI Video Dubber does

It downloads your file, pulls the audio out and sends it for transcription with timestamps, so every line of speech comes back with a start and an end.

Those lines are translated in one pass, which keeps them consistent with each other rather than translating each in isolation. Then each translated line is spoken separately and dropped back at the timestamp its original sat on, so the dub follows the picture instead of drifting.

Where a translated line runs longer than the gap it has to fit into, it is sped up to make room, up to three times. Past that it will not compress any further, and the next line starts while it is still speaking.

23 languages: Spanish, French, German, Italian, Portuguese, Hindi, Arabic, Russian, Japanese, Korean, Chinese, Dutch, Polish, Turkish, Indonesian, Vietnamese, Thai, Swedish, Ukrainian, Romanian, Greek, Hebrew and Filipino.

📥 What you give it

{
  "videoUrl": "https://yourdomain.com/clips/interview.mp4",
  "targetLanguage": "es",
  "voice": "onyx",
  "burnSubtitles": true
}

Field	Default	What it is
videoUrl	none	A public direct link to an .mp4, .mov or .webm. You host the file. The form prefills an example.com address, which is a placeholder, not a default: leave it there with a key set and the run fails on a missing file.
targetLanguage	es	The language to dub into, as a two-letter code from the list above.
sourceLanguage	auto	The spoken language of the original. Leave it on auto unless detection gets it wrong.
voice	onyx	alloy, echo, fable, onyx, nova or shimmer. onyx is the deep male one, nova and shimmer are female.
burnSubtitles	true	Overlays the translated subtitles on the picture. This means re-encoding the video, so it is slower. Turn it off and the picture is copied untouched.
openaiApiKey	none	Your own OpenAI key, used for all three steps. Marked secret.
ttsModel	tts-1	tts-1 is quick, tts-1-hd sounds better and costs you more on your own bill.
translationModel	gpt-4o-mini	The chat model that does the translating. Any model name that works on your host.
baseUrl	none	Advanced. Any OpenAI-compatible host. Empty means https://api.openai.com/v1.

Leave the key out and you get a labelled sample row rather than a dub, free, so you can see the shape before you spend anything.

📤 What you get back

One row, and two files in the run's key-value store: dubbed-es-1786411663909.mp4 and subtitles-es-1786411663909.srt, with the language code in the name.

Field	What it is
ok	true on a delivered dub.
videoUrl	The source link you gave it.
sourceLanguage	What the speech was detected as, or what you set.
targetLanguage, voice	What it was dubbed into, and who read it.
segments	How many separate lines of speech were found and voiced.
durationSeconds	The length of the source video, rounded. This is what the charge is worked out from.
output	mp4Key, srtKey and mp4Url: where the dubbed video and the subtitle file ended up.
processingSeconds	How long the run took.

No example row is printed here. No run on record has produced one with a working customer key, and a plausible row typed out by hand is worse than none.

🧾 Reading the output

Two rows can appear, and they look alike in the table, which is the one thing to watch.

Row	How to spot it	Billed
A finished dub	output.mp4Key has a filename in it	yes
The keyless sample	_sample: true, a reason, and output keys that are empty	no

The sample row carries ok: true, so the Overview table will not tell you apart from a real result. Check _sample or output.mp4Key before you trust a row.

If anything in the chain fails, the run stops and is marked failed. There is no partial row and no charge for a dub that was never made. The log names the step: the download, the transcription, the translation or the voice.

One more thing to check on the way out. When the translation comes back with fewer lines than there were segments, the gaps are filled with the original text, which means those lines get spoken in the source language by the dubbing voice. Listen through before you publish.

In the Overview table, output shows as a nested object rather than a link, so open the row or go to the run's Storage tab to get the file.

▶️ How to run it

1. Put your video somewhere with a direct download link. This does not fetch from video sites.
2. Open AI Video Dubber and click Try for free to see the sample row.
3. Paste your key into OpenAI API key (BYO) and your link into Video URL.
4. Pick Target language and Voice, and decide whether you want subtitles burned on.
5. Click Start, then take the MP4 and SRT from the run's Storage tab.

💰 How much does it cost?

$0.12 per minute of source video. Flat on every Apify plan, no volume tiers. Minutes are rounded up, so a 90-second clip counts as two and anything under a minute counts as one.

A run that fails delivers no video and is not charged for one. OpenAI bills your own key separately for the transcription, the translation and every line of speech, and tts-1-hd costs you more there than tts-1.

💡 What people use it for

• Putting an existing channel's back catalogue into a second language without re-recording anything.
• Course and training video localisation, where the SRT matters as much as the audio.
• Ads and product clips that need a Spanish or Arabic cut for one campaign.
• Getting a rough dub in front of a client to decide whether a proper voice session is worth booking.

🚧 What it does not do

• No voice cloning and no lip sync. One of six stock voices reads the whole thing, and mouths will not match.
• The original audio is replaced, not mixed underneath. Music and effects that were on the original track go with it.
• Keep sources short. The whole soundtrack goes up for transcription in one piece, so anything past roughly 13 minutes of audio is likely to fail at that step.
• One language and one video per run. Dubbing into three languages is three runs.
• It does not fetch from video sites. A direct file link to something you host, nothing else.
• A line that will not fit can overlap the next one. Speed-fitting stops at three times, and after that two voices can be heard together for a moment.
• No speech, no dub. A video with no audio track, or with nothing spoken in it, stops the run.

🧭 Which AI video actor do you need?

If you want	Use
A video re-voiced in another language	This one
Subtitles translated without touching the audio	Subtitle Translator
A plain transcript of what was said	Video Audio Transcriber
Word-by-word captions burned on your own video	Auto Caption Burner
A voiceover from text, with no video involved	AI Text-to-Speech Voiceover

❓ Questions people ask

Do I need my own key? Yes. Without it you get the sample row rather than a dub.

Can I keep the original audio underneath? No. The dub replaces the audio track.

Will it sound like the original speaker? No. It is one of six stock voices, picked by you.

How long does a run take? It depends on how many lines of speech there are, since each is voiced one at a time, and burning subtitles adds a full re-encode on top.

What do the rounded minutes mean for a short clip? A 20-second clip is charged as one minute, and a 3-minute-10-second one as four.

Can I dub a YouTube link? Not directly. Download the file first and host it somewhere with a direct link.

🆘 If something breaks

Open the Issues tab on the actor page. Include the run ID and your source link, and say which step the log stopped at, since that usually points straight at the cause.